shoichi-hasegawa0628/ldas

Implementation of LDAs

Implementation of LDAs

This repository is an implementation of LDA.
Implementation contents are as follows　　

1. Latent Dirichlet Allocation as a probabilistic generative model.
2. Multimodal Dirichlet Allocation as a probabilistic generative model.

Latent Dirichlet Allocation

Original paper

What the LDA contains:

• /LDA/lda.py:LDA by using Collapsed Gibbs sampler with Python.You need to decide the number of topic.

def main():
    n = 100 # データの水増し用の変数
    topic = 3 # トピック数を指定
    data = np.loadtxt( root , dtype=np.int32)*n # 発生回数にnをかけて水増し可能
    #print(data)
    lda( data , topic )

※Before running this script, you need to run /LDA/bow/bow.py which creates a BoW file

• /LDA/bow/bow.py: It can generate BoW file which is used by /LDA/lda.py from /LDA/bow/text.txt
• /LDA/bow/text.txt: You can write sentences.
  ※Sentences must be separated by spaces for each word.
  • For example

  内藤 は 彼女 が できない
  内藤 は 理想 が 高すぎる
  あの 店 の ラーメン は おいしい
  おいしい 店 の ごはん を 食べる
  室 は ずっと 踊って いる
  室 は 顔芸 を して いる
  

Usage:

1. Write sentences in /LDA/bow/text.txt
2. Create BoW:python3 /LDA/bow/bow.py
3. python3 /LDA/lda.py

Requirement
Python3.X and the following modules are required

import codecs
import numpy as np
from scipy.sparse import load_npz
import random
import math
import matplotlib.pyplot as plt

References:

• Implementation source of LDA

Multimodal Latent Dirichlet Allocation

What the MLDA contains:

• /MLDA/mlda.py:MLDA by using Collapsed Gibbs sampler with Python.You need to decide the number of topic.

def main():
    topic = 3
    data = []
    data.append( np.loadtxt( "./bof/histogram_v.txt" , dtype=np.int32) )
    data.append( np.loadtxt( "./bow/histogram_w.txt" , dtype=np.int32)*5 )
    mlda( data, topic, 100, "learn_result" )

    data[1] = None
    mlda( data, topic, 10, "recog_result" , "learn_result" )

※Before running this script, you need to run /MLDA/bow/bow.py and /MLDA/bof/bof.py which create BoW and BoF file

• /MLDA/bow/bow.py: It can generate BoW file which is used by /MLDA/mlda.py from /MLDA/bow/text.txt

• /MLDA/bow/text.txt: You can write sentences.
  ※Sentences must be separated by spaces for each word.

  • For example

  内藤 は 彼女 が できない
  内藤 は 理想 が 高すぎる
  あの 店 の ラーメン は おいしい
  おいしい 店 の ごはん を 食べる
  室 は ずっと 踊って いる
  室 は 顔芸 を して いる
  

• /MLDA/bof/bof.py: It can generate BoF file which is used by /MLDA/mlda.py from /MLDA/bof/images/*.png

Usage:

1. Write sentences in /MLDA/bow/text.txt
2. Create BoW:python3 /MLDA/bow/bow.py
3. Create BoF:python3 /MLDA/bof/bof.py
4. python3 /MLDA/mlda.py

Requirement:

Python3.X and the following modules are required

import numpy as np
import random
import math
import pylab
import pickle
import os
import sys
import cv2
import glob
import codecs

References:

• Implementation source of MLDA